Special Session

Special Session

Artificial Intelligence & Machine Learning

Fundamental Problems in Controller Design: Model Uncertainty

Dr. Phil Kim
J .MARPLE, Ltd., Korea

Oct. 15 (Thursday) 15:10~15:40 / Auditorium 1

System models are very important in controller design. It is no exaggeration to say that controller performance depends entirely on the accuracy of the model. Many control laws have been developed to solve this problem, but PID controller is still overwhelming in the field. In this talk, we will take a look at why PID controllers are still very popular in terms of response to model uncertainty. Based on this, we discuss the pros and cons of modern controllers. In particular, we will discuss how data-based AI technology can respond to model uncertainty and introduce some typical examples.


Neural sequence modeling – Foundation, Success and Questions

Prof. Kyunghyun Cho
New York University, USA

Oct. 15 (Thursday) 15:40~16:10 / Auditorium 1

In this talk, I will go over the foundations upon which neural sequence modeling is built. This foundation consists of autoregressive density modeling, latent variable models and powerful neural net based parametrization. This explanation of the foundations will be naturally followed by the success of neural sequence modeling in various domains, including machine translation and its application to question-answering-based summary evaluation. These success stories however are not without any caveats, and I will conclude this talk by pointing out and discussing questions that should have been asked in 2014 but have not been answered even now.